https://gcc.gnu.org/bugzilla/show_bug.cgi?id=3507
Joseph C. Sible <josephcsible at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |josephcsible at gmail dot com
--- Comment #59 from Joseph C. Sible <josephcsible at gmail dot com> ---
x86_64-*-* should be added to the target list. With the original test case, GCC
10.1 produces this:
foo:
movq %rdi, %rax
subq %rsi, %rax
cmpq %rsi, %rdi
leaq 100(%rax), %rdx
cmovb %rdx, %rax
ret
It should instead produce this:
foo:
subq %rsi, %rdi
leaq 100(%rdi), %rax
cmovae %rdi, %rax
ret
By the way, I thought that using __builtin_usubl_overflow might be a useful
workaround for this bug, but it turns out that it currently isn't really, since
it has a different missed optimization of its own: bug #96289
